Mindful Clinical Copilot
Mindful Clinical Copilot Overview
Overview:
Mindful Clinical Courses is our flagship AI platform designed to support healthcare students and professionals across the UK in their academic and career progression. It offers personalized, curriculum-specific tutoring, exam preparation, and group study tools powered by multi-agent AI.
Problem We Solve:
Healthcare learners often pay separately for question banks, masterclasses, and revision tools — many of which are expensive, fragmented, and not personalized. Our platform aims to disrupt this model by offering an all-in-one, AI-powered solution at a minimal cost or free via NHS or university partnerships.
Key Features:
-
MCQ Practice: AI explanations, mistake review, mock exams
-
Lecture Study Hub: AI-generated notes from official resources, flashcards, illustrations, and videos
-
OSCE Hub: Real-time voice-based patient simulations using.
-
Group Study: AI-moderated collaborative sessions with scheduling and summaries.
-
University Hub: Course-specific tutoring grounded in UK guidelines
-
Multi-AI Architecture:
-
Gemini & GPT for tutoring and conversation
-
ElevenLabs for voice narration
-
On-demand illustration and animation generation for study aids from RAG data..
-
Target Audience:
-
UK medical, nursing, paramedic, pharmacy, and allied health students
-
NHS staff preparing for postgraduate exams
-
Universities and training providers seeking scalable AI support
Live Preview:
🌐 https://mindfulclinicalcourses.co.uk
Please note that this is where most parts of the Mindful Clinical Copilot will initially be launched for specialist healthcare and university healthcare learning.
Mindful Clinical Copilot (MCC)
A safe, grounded, and affordable AI platform for clinical education, early-help, and service navigation in the UK.
Project Maturity: Multi-module platform; stabilization complete; private pilots underway; demo-ready tracks
1. Executive Summary
Mindful Clinical Copilot (MCC) is an AI-native learning and early-help platform that turns static content into a grounded, interactive, and governance-aware experience. It is not a single "feature"—like a simple notetaker—but a scalable platform of integrated modules for MCQ practice, lecture study, OSCE simulations, group study, and deep-document learning. Our architecture is built from the ground up on enterprise-grade, secure Google Cloud infrastructure, ensuring data privacy and compliance are at its core.
Our core mission is to solve the two biggest problems in digital education: AI hallucination and learner error. We do this with two unique innovations:
-
A 4-Tier "Grounded & Reconciled" RAG System: Our AI is not a generic chatbot. It operates on a strict, 4-tier fallback system that prioritizes our own curated, expert-vetted data. Before discussion, it validates this data against live national guidelines (NICE, BNF) using Google's verifiable Vertex AI Search, providing a traceable, "un-hallucinated" answer every time.
-
"Custom Study" Reconciliation: We are the only platform that solves the "garbage in, garbage out" problem. When a student uploads their own notes, our AI first compares them against our RAG system and flags any outdated or incorrect statements before the tutoring session begins, preventing the reinforcement of misinformation.
We are seeking funding to finish the last-mile unification of our platform (voice+text parity, Trust-level RAG), scale our B2B pilots with universities and NHS Trusts, and launch an affordable subscription for all UK healthcare learners.
2. The Problem: Hallucinations, "Crash Courses," and "Feature-Gimmicks"
The market for AI in medical education is crowded, but it is failing our future clinicians in three critical ways:
-
The "AI Hallucination" Crisis: The market is saturated with generic AI chatbots that claim to cite data. This is a black box. In medicine, a plausible-sounding hallucination is a direct threat to patient safety. Furthermore, these AIs will happily "discuss" a user's uploaded documents, but what if those documents are wrong? If a student uploads last year's outdated notes, the AI will confidently "teach" them last year's wrong information, reinforcing dangerous habits.
-
The Inefficient "Crash Course" Model: Busy clinicians and trainees spend thousands of pounds on inconvenient, exhausting "crash courses" to pass Royal College exams. These are often taught by recently-passed peers who, despite their best intentions, are not professional educators. They may be tired, "can't recall everything," and inadvertently fill in gaps with untrue information—a human form of hallucination.
-
"Features" Pretending to be Platforms: Many "flagship" AI products are just single-feature gimmicks—like a simple scribe or document-chatter—masquerading as a scalable solution. This creates a fragmented, expensive, and non-integrated experience for the learner.
3. The MCC Solution: A Platform Built on Trust
MCC is not a feature; it is a scalable platform that solves these problems with three core principles.
3.1. Safety via our 4-Tier "Grounded & Reconciled" RAG
Our AI Tutor is not a generic chatbot. It operates on a 4-tier fallback system to ensure every answer is safe and verifiable.
-
Tier 1: Curated Content (Our Truth): The AI first pulls from our own expert-vetted, static data files. This is our single source of truth.
-
Tier 2: Live Validation: Before using Tier 1 data, the AI performs a quick check against live guidelines (Tier 3) to ensure the curated answer isn't outdated.
-
Tier 3: Institutional & Web RAG: If the topic isn't in our local data, the AI performs a site-restricted search on pre-approved UK sources (NICE, BNF, NHS) and specific Trust data using Google Cloud's enterprise-grade Vertex AI Search. This allows us to create a private, indexed search over trusted documents, ensuring that our AI's knowledge is not just grounded, but also auditable and securely managed within a trusted cloud environment.
-
Tier 4: Guardrailed LLM: Only as a last resort does the AI use its general knowledge, with strict guardrails to prevent it from providing clinical advice outside its grounded context.
3.2. The "Always-On" Tutor (Solving the Crash Course)
This is our solution to the expensive, inconvenient "crash course" model.
-
AI Moderator: A group can start a study session at any time, day or night. The AI Moderator facilitates, poses questions, and can save and recap the session.
-
"Passive Learning Podcast": This is our killer feature. The AI Moderator can run an automated "Masterclass." It will narrate a lecture topic (using TTS) while simultaneously auto-navigating the shared-screen content.
-
Audio-First Learning: This "podcast" mode allows users to learn passively while commuting. They can even interact with it using a wake word (e.g., "Hey Copilot, skip to the next question").
3.3. A True Platform (Not Just Features)
Our platform is a single, unified project that can be broken down and delivered as modular products. It's an ecosystem, not a single tool.
-
MCQ, OSCE, and Lecture Hubs: The core learner modules for knowledge and skills.
-
Custom Study "Notebooks": This is where we solve the "garbage in, garbage out" problem. Users can upload their notes, videos, and lecture recordings. Our AI reconciles this content against our RAG system, flags errors, and then allows the user to chat and generate study aids from the validated information.
-
The University Hub: This is a major, overlooked pillar. We have researched and mapped all UK-based healthcare courses. This allows us to provide a unified platform for all healthcare students and NHS employees, giving everyone an equal opportunity to learn what they need. It is our primary B2B strategy, aiming for collaboration with schools for unified, free-at-the-point-of-use accounts.
4. Enterprise-Grade Security, Compliance, and Data Privacy
Trust is not just about accurate information; it's about safeguarding user data. Our platform is architected with security and privacy as foundational pillars, not afterthoughts.
Secure Infrastructure with Google Firebase: Our entire backend infrastructure is built on Google Firebase, a platform renowned for its robust, multi-layered security built on Google's world-class cloud infrastructure. We leverage:
-
Firebase Authentication: For secure, scalable user identity management, protecting user credentials while supporting various sign-in methods.
-
Firestore Security Rules: To ensure granular, server-side access control, guaranteeing that users can only access their own data. All data is encrypted both in
Project Demo
NHS Collaboration
BSMHFT Demo
Project Title & Tagline
BSMHFT Copilot – “Empowering Care with Safe, Grounded AI”
Overview
The Mindful Clinical Copilot is a governed, role-aware AI platform designed for Birmingham and Solihull Mental Health NHS Foundation Trust (BSMHFT). It reduces documentation burden, improves access to verified guidance, and enhances patient safety through grounded AI responses.
Modules:
• Clinician AI Scribe – Live transcription and template-driven note drafting.
• Staff Assistant – Quick access to Trust policies, training catalogues, and wellbeing resources.
• Patient Companion – Empathetic, supervised chatbot with crisis escalation and verified resources.
Key Features
• Interpreter Mode (Gemini Live) – Two-way interpretation for common UK languages during clinical interviews.
• Moderator Dashboard – Transcript verification, audit trail, and anonymised KPI export.
• Governance-first design – DPIA, DCB0160 safety case, MHRA non-SaMD stance, DTAC compliance.
QI Project Summary
A 12-week Quality Improvement pilot approved by BSMHFT leadership:
• Aim: Reduce documentation time by ≥30%, achieve ≥90% policy answer accuracy, ≥95% crisis detection correctness.
• Scope: Two inpatient wards + one CMHT; supervised patient use; bilingual moderator review.
• KPIs: Time saved per note, interpreter accuracy, staff wellbeing improvements, usability SUS ≥70.
Technology & Compliance
• Vertex AI Search – europe-west2 (London) for public RAG sources.
• Firebase – EU multi-region for authentication, Firestore, and Storage.
• NHSmail SSO – Microsoft Entra ID integration with allow-list and MFA.
• Retention – Audio recordings deleted after 30 days; transcripts anonymised post-QI.
• Compliance – DTAC, DPIA (NHS template), DCB0129/DCB0160, DSP Toolkit alignment.
Impact & Benefits
• Time savings: Clinicians reclaim up to 20 minutes per patient note.
• Interpreter cost reduction: AI Interpreter Mode offsets 25–40% of low-risk sessions.
• Staff wellbeing: Faster access to resources and reduced stress scores.
• Patient safety: Crisis triggers handled with deterministic guardrails and immediate escalation.
Future Vision
• Enterprise rollout across BSMHFT with NHSmail SSO integration and Azure Blob indexing for intranet policies.
• National adoption via NHS frameworks (G-Cloud, SBS) with DTAC and clinical safety packs.
• Expansion into interpreter services, community teams, and HTT for measurable cost savings and capacity gains.
Call to Action for Partners & Funders
Join us in scaling a safe, grounded AI solution that improves care quality and efficiency. We welcome collaboration and funding support through Google Cloud credits, Microsoft for Nonprofits grants, and NHS innovation partnerships.
